档案资源开发2026年第40卷第4期《档案学研究》
基于混合深度学习的满族档案知识重组研究——以《满文老档》汉译本为例
Research on Knowledge Reorganization of Manchu Archives Based on Hybrid Deep Learning: A Case Study of the Chinese Translation of Old Manchu Archives
东北师范大学信息科学与技术学院 长春 130117
摘要
针对满族档案知识组织与开发的不足,以《满文老档》为语料来源,结合混合深度学习技术与大语言模型,分别探索《满文老档》满文本知识元识别的可行性、《满文老档》汉译本的知识元识别效果及《满文老档》中复杂语义知识元的抽取。在此基础上,基于知识元抽取结果探索满族档案知识元的重组利用。研究发现:直接开展满文本知识元识别的可行性较差,需要探索面向满文的预训练模型构建;GuwenBERT-BiLSTM-CRF模型在满族档案知识元识别方面效果良好,且在识别职官类和时间类知识元上效果最优;大语言模型在复杂语义知识元识别上效果尚可,但仍需人工进行二次处理以保障质量;满族档案知识图谱能够实现满族档案知识关联、知识检索与知识推理,推动满族档案知识的保护、开发和利用。
关键词:满族档案知识元知识图谱命名实体识别
Abstract
In view of the deficiencies in knowledge organization and development of Manchu archives, the Old Manchu Archives were selected as the corpus source. By integrating hybrid deep learning techniques with large language models, this study explores the the feasibility of identifying knowledge elements in the Manchu text of the Old Manchu Archives, the effectiveness of knowledge element identification in the Chinese translation of the Old Manchu Archives, and the extraction of complex semantic knowledge elements within the Old Manchu Archives. Building upon these findings, the study further investigated the reorganization and utilization of knowledge elements from the Manchu archives based on the results of knowledge element identification.The study found: Directly performing knowledge entity recognition on Manchu text is not feasible, it is necessary to explore the construction of pre-trained models tailored for Manchu text; GuwenBERT-BiLSTM-CRF model has a good effect on knowledge element recognition of Manchu archives, and has the best effect on official position and time categories; Large language model approaches demonstrate acceptable performance in identifying complex semantic knowledge elements, but secondary manual processing remains necessary to obtain high-quality knowledge elements; The knowledge graph of Manchu archives can realize the knowledge association, knowledge retrieval and knowledge reasoning of Manchu archives, advancing the preservation, development, and utilization of the knowledge of Manchu archives.
Key words: Manchu archives; knowledge elements; knowledge graph; named entity recognition
引用格式
参考文献


